Applications of artificial intelligence in oncology patient education: A scoping review.

e13642 Background: As cancer care becomes increasingly complex, artificial intelligence (AI) tools have the potential to improve patient understanding, personalize communication, and support shared decision-making. Reviewing current literature on these applications is crucial to identifying effective strategies and addressing gaps in implementing AI-driven interventions in oncology care. Methods: We searched PubMed, EMBASE, Cochrane, Medline, and World of Science from January 2015 through November 2024 for original studies on implementations of AI in patient-facing oncology education. Abstracts and short communications were excluded. Data extraction included date of publication, cancer type(s), AI model type(s), study design, purpose, and findings. Reported score metrics included accuracy, comprehensibility, and readability of AI-generated education materials. Results: 1,437 publications were identified based on search terms, of which 77 studies were included. 59/77 (77%) studies were published in 2024. 13/77 (17%) focused on general oncology education, including symptom/pain control and treatment toxicities. 64/77 (83%) focused on addressing patient inquiries within specific cancer types, with prostate (20), breast (16), and colorectal (8) being the most common. 72/77 studies (94%) utilized some form of large language model (LLM) to summarize and deliver education materials, most of which used publicly available chatbots such as ChatGPT 3.5/4.0, Google Gemini, and Microsoft Copilot. The remaining studies used AI to build interactive platforms, virtual coaching assistants, and educational video generators. The role of AI in patient education in oncology is currently limited to (1) responding to patient questions, and (2) simplifying existing medical guidelines and texts to be more patient-friendly. Congregated data shows that AI-produced information has an overall accuracy of 81% (IQR 75%-87%) and comprehensibility of 83% (IQR 75%-90%). 21 studies used AI to improve readability of patient education materials, achieving an overall reduction in reading level from “post-college grad” to “high-school” (based on Flesch-Kincaid Grading). Only 2 studies achieved AMA and NIH’s recommendation of sixth-grade reading level. Conclusions: AI introduced novel and creative approaches to delivering oncology information to patients, although caution is needed due to limitations in accuracy and comprehensibility. In particular, AI-enhanced readability may still exceed the recommended level for patient education materials, highlighting the need for further optimization. Future research will be essential to refining AI tools and maximizing their clinical impact.